projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
fuse: use req->page_descs[] for argpages cases
[deliverable/linux.git] / fs / fuse / file.c
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7 */
8
9 #include "fuse_i.h"
10
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/kernel.h>
14 #include <linux/sched.h>
15 #include <linux/module.h>
16 #include <linux/compat.h>
17 #include <linux/swap.h>
18
19 static const struct file_operations fuse_direct_io_file_operations;
20
21 static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
22 int opcode, struct fuse_open_out *outargp)
23 {
24 struct fuse_open_in inarg;
25 struct fuse_req *req;
26 int err;
27
28 req = fuse_get_req_nopages(fc);
29 if (IS_ERR(req))
30 return PTR_ERR(req);
31
32 memset(&inarg, 0, sizeof(inarg));
33 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
34 if (!fc->atomic_o_trunc)
35 inarg.flags &= ~O_TRUNC;
36 req->in.h.opcode = opcode;
37 req->in.h.nodeid = nodeid;
38 req->in.numargs = 1;
39 req->in.args[0].size = sizeof(inarg);
40 req->in.args[0].value = &inarg;
41 req->out.numargs = 1;
42 req->out.args[0].size = sizeof(*outargp);
43 req->out.args[0].value = outargp;
44 fuse_request_send(fc, req);
45 err = req->out.h.error;
46 fuse_put_request(fc, req);
47
48 return err;
49 }
50
51 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
52 {
53 struct fuse_file *ff;
54
55 ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
56 if (unlikely(!ff))
57 return NULL;
58
59 ff->fc = fc;
60 ff->reserved_req = fuse_request_alloc(0);
61 if (unlikely(!ff->reserved_req)) {
62 kfree(ff);
63 return NULL;
64 }
65
66 INIT_LIST_HEAD(&ff->write_entry);
67 atomic_set(&ff->count, 0);
68 RB_CLEAR_NODE(&ff->polled_node);
69 init_waitqueue_head(&ff->poll_wait);
70
71 spin_lock(&fc->lock);
72 ff->kh = ++fc->khctr;
73 spin_unlock(&fc->lock);
74
75 return ff;
76 }
77
78 void fuse_file_free(struct fuse_file *ff)
79 {
80 fuse_request_free(ff->reserved_req);
81 kfree(ff);
82 }
83
84 struct fuse_file *fuse_file_get(struct fuse_file *ff)
85 {
86 atomic_inc(&ff->count);
87 return ff;
88 }
89
90 static void fuse_release_async(struct work_struct *work)
91 {
92 struct fuse_req *req;
93 struct fuse_conn *fc;
94 struct path path;
95
96 req = container_of(work, struct fuse_req, misc.release.work);
97 path = req->misc.release.path;
98 fc = get_fuse_conn(path.dentry->d_inode);
99
100 fuse_put_request(fc, req);
101 path_put(&path);
102 }
103
104 static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
105 {
106 if (fc->destroy_req) {
107 /*
108 * If this is a fuseblk mount, then it's possible that
109 * releasing the path will result in releasing the
110 * super block and sending the DESTROY request. If
111 * the server is single threaded, this would hang.
112 * For this reason do the path_put() in a separate
113 * thread.
114 */
115 atomic_inc(&req->count);
116 INIT_WORK(&req->misc.release.work, fuse_release_async);
117 schedule_work(&req->misc.release.work);
118 } else {
119 path_put(&req->misc.release.path);
120 }
121 }
122
123 static void fuse_file_put(struct fuse_file *ff, bool sync)
124 {
125 if (atomic_dec_and_test(&ff->count)) {
126 struct fuse_req *req = ff->reserved_req;
127
128 if (sync) {
129 fuse_request_send(ff->fc, req);
130 path_put(&req->misc.release.path);
131 fuse_put_request(ff->fc, req);
132 } else {
133 req->end = fuse_release_end;
134 fuse_request_send_background(ff->fc, req);
135 }
136 kfree(ff);
137 }
138 }
139
140 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
141 bool isdir)
142 {
143 struct fuse_open_out outarg;
144 struct fuse_file *ff;
145 int err;
146 int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
147
148 ff = fuse_file_alloc(fc);
149 if (!ff)
150 return -ENOMEM;
151
152 err = fuse_send_open(fc, nodeid, file, opcode, &outarg);
153 if (err) {
154 fuse_file_free(ff);
155 return err;
156 }
157
158 if (isdir)
159 outarg.open_flags &= ~FOPEN_DIRECT_IO;
160
161 ff->fh = outarg.fh;
162 ff->nodeid = nodeid;
163 ff->open_flags = outarg.open_flags;
164 file->private_data = fuse_file_get(ff);
165
166 return 0;
167 }
168 EXPORT_SYMBOL_GPL(fuse_do_open);
169
170 void fuse_finish_open(struct inode *inode, struct file *file)
171 {
172 struct fuse_file *ff = file->private_data;
173 struct fuse_conn *fc = get_fuse_conn(inode);
174
175 if (ff->open_flags & FOPEN_DIRECT_IO)
176 file->f_op = &fuse_direct_io_file_operations;
177 if (!(ff->open_flags & FOPEN_KEEP_CACHE))
178 invalidate_inode_pages2(inode->i_mapping);
179 if (ff->open_flags & FOPEN_NONSEEKABLE)
180 nonseekable_open(inode, file);
181 if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
182 struct fuse_inode *fi = get_fuse_inode(inode);
183
184 spin_lock(&fc->lock);
185 fi->attr_version = ++fc->attr_version;
186 i_size_write(inode, 0);
187 spin_unlock(&fc->lock);
188 fuse_invalidate_attr(inode);
189 }
190 }
191
192 int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
193 {
194 struct fuse_conn *fc = get_fuse_conn(inode);
195 int err;
196
197 err = generic_file_open(inode, file);
198 if (err)
199 return err;
200
201 err = fuse_do_open(fc, get_node_id(inode), file, isdir);
202 if (err)
203 return err;
204
205 fuse_finish_open(inode, file);
206
207 return 0;
208 }
209
210 static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
211 {
212 struct fuse_conn *fc = ff->fc;
213 struct fuse_req *req = ff->reserved_req;
214 struct fuse_release_in *inarg = &req->misc.release.in;
215
216 spin_lock(&fc->lock);
217 list_del(&ff->write_entry);
218 if (!RB_EMPTY_NODE(&ff->polled_node))
219 rb_erase(&ff->polled_node, &fc->polled_files);
220 spin_unlock(&fc->lock);
221
222 wake_up_interruptible_all(&ff->poll_wait);
223
224 inarg->fh = ff->fh;
225 inarg->flags = flags;
226 req->in.h.opcode = opcode;
227 req->in.h.nodeid = ff->nodeid;
228 req->in.numargs = 1;
229 req->in.args[0].size = sizeof(struct fuse_release_in);
230 req->in.args[0].value = inarg;
231 }
232
233 void fuse_release_common(struct file *file, int opcode)
234 {
235 struct fuse_file *ff;
236 struct fuse_req *req;
237
238 ff = file->private_data;
239 if (unlikely(!ff))
240 return;
241
242 req = ff->reserved_req;
243 fuse_prepare_release(ff, file->f_flags, opcode);
244
245 if (ff->flock) {
246 struct fuse_release_in *inarg = &req->misc.release.in;
247 inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK;
248 inarg->lock_owner = fuse_lock_owner_id(ff->fc,
249 (fl_owner_t) file);
250 }
251 /* Hold vfsmount and dentry until release is finished */
252 path_get(&file->f_path);
253 req->misc.release.path = file->f_path;
254
255 /*
256 * Normally this will send the RELEASE request, however if
257 * some asynchronous READ or WRITE requests are outstanding,
258 * the sending will be delayed.
259 *
260 * Make the release synchronous if this is a fuseblk mount,
261 * synchronous RELEASE is allowed (and desirable) in this case
262 * because the server can be trusted not to screw up.
263 */
264 fuse_file_put(ff, ff->fc->destroy_req != NULL);
265 }
266
267 static int fuse_open(struct inode *inode, struct file *file)
268 {
269 return fuse_open_common(inode, file, false);
270 }
271
272 static int fuse_release(struct inode *inode, struct file *file)
273 {
274 fuse_release_common(file, FUSE_RELEASE);
275
276 /* return value is ignored by VFS */
277 return 0;
278 }
279
280 void fuse_sync_release(struct fuse_file *ff, int flags)
281 {
282 WARN_ON(atomic_read(&ff->count) > 1);
283 fuse_prepare_release(ff, flags, FUSE_RELEASE);
284 ff->reserved_req->force = 1;
285 fuse_request_send(ff->fc, ff->reserved_req);
286 fuse_put_request(ff->fc, ff->reserved_req);
287 kfree(ff);
288 }
289 EXPORT_SYMBOL_GPL(fuse_sync_release);
290
291 /*
292 * Scramble the ID space with XTEA, so that the value of the files_struct
293 * pointer is not exposed to userspace.
294 */
295 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
296 {
297 u32 *k = fc->scramble_key;
298 u64 v = (unsigned long) id;
299 u32 v0 = v;
300 u32 v1 = v >> 32;
301 u32 sum = 0;
302 int i;
303
304 for (i = 0; i < 32; i++) {
305 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
306 sum += 0x9E3779B9;
307 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
308 }
309
310 return (u64) v0 + ((u64) v1 << 32);
311 }
312
313 /*
314 * Check if page is under writeback
315 *
316 * This is currently done by walking the list of writepage requests
317 * for the inode, which can be pretty inefficient.
318 */
319 static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
320 {
321 struct fuse_conn *fc = get_fuse_conn(inode);
322 struct fuse_inode *fi = get_fuse_inode(inode);
323 struct fuse_req *req;
324 bool found = false;
325
326 spin_lock(&fc->lock);
327 list_for_each_entry(req, &fi->writepages, writepages_entry) {
328 pgoff_t curr_index;
329
330 BUG_ON(req->inode != inode);
331 curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT;
332 if (curr_index == index) {
333 found = true;
334 break;
335 }
336 }
337 spin_unlock(&fc->lock);
338
339 return found;
340 }
341
342 /*
343 * Wait for page writeback to be completed.
344 *
345 * Since fuse doesn't rely on the VM writeback tracking, this has to
346 * use some other means.
347 */
348 static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
349 {
350 struct fuse_inode *fi = get_fuse_inode(inode);
351
352 wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
353 return 0;
354 }
355
356 static int fuse_flush(struct file *file, fl_owner_t id)
357 {
358 struct inode *inode = file->f_path.dentry->d_inode;
359 struct fuse_conn *fc = get_fuse_conn(inode);
360 struct fuse_file *ff = file->private_data;
361 struct fuse_req *req;
362 struct fuse_flush_in inarg;
363 int err;
364
365 if (is_bad_inode(inode))
366 return -EIO;
367
368 if (fc->no_flush)
369 return 0;
370
371 req = fuse_get_req_nofail_nopages(fc, file);
372 memset(&inarg, 0, sizeof(inarg));
373 inarg.fh = ff->fh;
374 inarg.lock_owner = fuse_lock_owner_id(fc, id);
375 req->in.h.opcode = FUSE_FLUSH;
376 req->in.h.nodeid = get_node_id(inode);
377 req->in.numargs = 1;
378 req->in.args[0].size = sizeof(inarg);
379 req->in.args[0].value = &inarg;
380 req->force = 1;
381 fuse_request_send(fc, req);
382 err = req->out.h.error;
383 fuse_put_request(fc, req);
384 if (err == -ENOSYS) {
385 fc->no_flush = 1;
386 err = 0;
387 }
388 return err;
389 }
390
391 /*
392 * Wait for all pending writepages on the inode to finish.
393 *
394 * This is currently done by blocking further writes with FUSE_NOWRITE
395 * and waiting for all sent writes to complete.
396 *
397 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
398 * could conflict with truncation.
399 */
400 static void fuse_sync_writes(struct inode *inode)
401 {
402 fuse_set_nowrite(inode);
403 fuse_release_nowrite(inode);
404 }
405
406 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
407 int datasync, int isdir)
408 {
409 struct inode *inode = file->f_mapping->host;
410 struct fuse_conn *fc = get_fuse_conn(inode);
411 struct fuse_file *ff = file->private_data;
412 struct fuse_req *req;
413 struct fuse_fsync_in inarg;
414 int err;
415
416 if (is_bad_inode(inode))
417 return -EIO;
418
419 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
420 if (err)
421 return err;
422
423 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
424 return 0;
425
426 mutex_lock(&inode->i_mutex);
427
428 /*
429 * Start writeback against all dirty pages of the inode, then
430 * wait for all outstanding writes, before sending the FSYNC
431 * request.
432 */
433 err = write_inode_now(inode, 0);
434 if (err)
435 goto out;
436
437 fuse_sync_writes(inode);
438
439 req = fuse_get_req_nopages(fc);
440 if (IS_ERR(req)) {
441 err = PTR_ERR(req);
442 goto out;
443 }
444
445 memset(&inarg, 0, sizeof(inarg));
446 inarg.fh = ff->fh;
447 inarg.fsync_flags = datasync ? 1 : 0;
448 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
449 req->in.h.nodeid = get_node_id(inode);
450 req->in.numargs = 1;
451 req->in.args[0].size = sizeof(inarg);
452 req->in.args[0].value = &inarg;
453 fuse_request_send(fc, req);
454 err = req->out.h.error;
455 fuse_put_request(fc, req);
456 if (err == -ENOSYS) {
457 if (isdir)
458 fc->no_fsyncdir = 1;
459 else
460 fc->no_fsync = 1;
461 err = 0;
462 }
463 out:
464 mutex_unlock(&inode->i_mutex);
465 return err;
466 }
467
468 static int fuse_fsync(struct file *file, loff_t start, loff_t end,
469 int datasync)
470 {
471 return fuse_fsync_common(file, start, end, datasync, 0);
472 }
473
474 void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos,
475 size_t count, int opcode)
476 {
477 struct fuse_read_in *inarg = &req->misc.read.in;
478 struct fuse_file *ff = file->private_data;
479
480 inarg->fh = ff->fh;
481 inarg->offset = pos;
482 inarg->size = count;
483 inarg->flags = file->f_flags;
484 req->in.h.opcode = opcode;
485 req->in.h.nodeid = ff->nodeid;
486 req->in.numargs = 1;
487 req->in.args[0].size = sizeof(struct fuse_read_in);
488 req->in.args[0].value = inarg;
489 req->out.argvar = 1;
490 req->out.numargs = 1;
491 req->out.args[0].size = count;
492 }
493
494 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
495 loff_t pos, size_t count, fl_owner_t owner)
496 {
497 struct fuse_file *ff = file->private_data;
498 struct fuse_conn *fc = ff->fc;
499
500 fuse_read_fill(req, file, pos, count, FUSE_READ);
501 if (owner != NULL) {
502 struct fuse_read_in *inarg = &req->misc.read.in;
503
504 inarg->read_flags |= FUSE_READ_LOCKOWNER;
505 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
506 }
507 fuse_request_send(fc, req);
508 return req->out.args[0].size;
509 }
510
511 static void fuse_read_update_size(struct inode *inode, loff_t size,
512 u64 attr_ver)
513 {
514 struct fuse_conn *fc = get_fuse_conn(inode);
515 struct fuse_inode *fi = get_fuse_inode(inode);
516
517 spin_lock(&fc->lock);
518 if (attr_ver == fi->attr_version && size < inode->i_size) {
519 fi->attr_version = ++fc->attr_version;
520 i_size_write(inode, size);
521 }
522 spin_unlock(&fc->lock);
523 }
524
525 static int fuse_readpage(struct file *file, struct page *page)
526 {
527 struct inode *inode = page->mapping->host;
528 struct fuse_conn *fc = get_fuse_conn(inode);
529 struct fuse_req *req;
530 size_t num_read;
531 loff_t pos = page_offset(page);
532 size_t count = PAGE_CACHE_SIZE;
533 u64 attr_ver;
534 int err;
535
536 err = -EIO;
537 if (is_bad_inode(inode))
538 goto out;
539
540 /*
541 * Page writeback can extend beyond the lifetime of the
542 * page-cache page, so make sure we read a properly synced
543 * page.
544 */
545 fuse_wait_on_page_writeback(inode, page->index);
546
547 req = fuse_get_req(fc, 1);
548 err = PTR_ERR(req);
549 if (IS_ERR(req))
550 goto out;
551
552 attr_ver = fuse_get_attr_version(fc);
553
554 req->out.page_zeroing = 1;
555 req->out.argpages = 1;
556 req->num_pages = 1;
557 req->pages[0] = page;
558 req->page_descs[0].length = count;
559 num_read = fuse_send_read(req, file, pos, count, NULL);
560 err = req->out.h.error;
561 fuse_put_request(fc, req);
562
563 if (!err) {
564 /*
565 * Short read means EOF. If file size is larger, truncate it
566 */
567 if (num_read < count)
568 fuse_read_update_size(inode, pos + num_read, attr_ver);
569
570 SetPageUptodate(page);
571 }
572
573 fuse_invalidate_attr(inode); /* atime changed */
574 out:
575 unlock_page(page);
576 return err;
577 }
578
579 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
580 {
581 int i;
582 size_t count = req->misc.read.in.size;
583 size_t num_read = req->out.args[0].size;
584 struct address_space *mapping = NULL;
585
586 for (i = 0; mapping == NULL && i < req->num_pages; i++)
587 mapping = req->pages[i]->mapping;
588
589 if (mapping) {
590 struct inode *inode = mapping->host;
591
592 /*
593 * Short read means EOF. If file size is larger, truncate it
594 */
595 if (!req->out.h.error && num_read < count) {
596 loff_t pos;
597
598 pos = page_offset(req->pages[0]) + num_read;
599 fuse_read_update_size(inode, pos,
600 req->misc.read.attr_ver);
601 }
602 fuse_invalidate_attr(inode); /* atime changed */
603 }
604
605 for (i = 0; i < req->num_pages; i++) {
606 struct page *page = req->pages[i];
607 if (!req->out.h.error)
608 SetPageUptodate(page);
609 else
610 SetPageError(page);
611 unlock_page(page);
612 page_cache_release(page);
613 }
614 if (req->ff)
615 fuse_file_put(req->ff, false);
616 }
617
618 static void fuse_send_readpages(struct fuse_req *req, struct file *file)
619 {
620 struct fuse_file *ff = file->private_data;
621 struct fuse_conn *fc = ff->fc;
622 loff_t pos = page_offset(req->pages[0]);
623 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
624
625 req->out.argpages = 1;
626 req->out.page_zeroing = 1;
627 req->out.page_replace = 1;
628 fuse_read_fill(req, file, pos, count, FUSE_READ);
629 req->misc.read.attr_ver = fuse_get_attr_version(fc);
630 if (fc->async_read) {
631 req->ff = fuse_file_get(ff);
632 req->end = fuse_readpages_end;
633 fuse_request_send_background(fc, req);
634 } else {
635 fuse_request_send(fc, req);
636 fuse_readpages_end(fc, req);
637 fuse_put_request(fc, req);
638 }
639 }
640
641 struct fuse_fill_data {
642 struct fuse_req *req;
643 struct file *file;
644 struct inode *inode;
645 unsigned nr_pages;
646 };
647
648 static int fuse_readpages_fill(void *_data, struct page *page)
649 {
650 struct fuse_fill_data *data = _data;
651 struct fuse_req *req = data->req;
652 struct inode *inode = data->inode;
653 struct fuse_conn *fc = get_fuse_conn(inode);
654
655 fuse_wait_on_page_writeback(inode, page->index);
656
657 if (req->num_pages &&
658 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
659 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
660 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
661 int nr_alloc = min_t(unsigned, data->nr_pages,
662 FUSE_MAX_PAGES_PER_REQ);
663 fuse_send_readpages(req, data->file);
664 data->req = req = fuse_get_req(fc, nr_alloc);
665 if (IS_ERR(req)) {
666 unlock_page(page);
667 return PTR_ERR(req);
668 }
669 }
670
671 if (WARN_ON(req->num_pages >= req->max_pages)) {
672 fuse_put_request(fc, req);
673 return -EIO;
674 }
675
676 page_cache_get(page);
677 req->pages[req->num_pages] = page;
678 req->page_descs[req->num_pages].length = PAGE_SIZE;
679 req->num_pages++;
680 data->nr_pages--;
681 return 0;
682 }
683
684 static int fuse_readpages(struct file *file, struct address_space *mapping,
685 struct list_head *pages, unsigned nr_pages)
686 {
687 struct inode *inode = mapping->host;
688 struct fuse_conn *fc = get_fuse_conn(inode);
689 struct fuse_fill_data data;
690 int err;
691 int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ);
692
693 err = -EIO;
694 if (is_bad_inode(inode))
695 goto out;
696
697 data.file = file;
698 data.inode = inode;
699 data.req = fuse_get_req(fc, nr_alloc);
700 data.nr_pages = nr_pages;
701 err = PTR_ERR(data.req);
702 if (IS_ERR(data.req))
703 goto out;
704
705 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
706 if (!err) {
707 if (data.req->num_pages)
708 fuse_send_readpages(data.req, file);
709 else
710 fuse_put_request(fc, data.req);
711 }
712 out:
713 return err;
714 }
715
716 static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
717 unsigned long nr_segs, loff_t pos)
718 {
719 struct inode *inode = iocb->ki_filp->f_mapping->host;
720 struct fuse_conn *fc = get_fuse_conn(inode);
721
722 /*
723 * In auto invalidate mode, always update attributes on read.
724 * Otherwise, only update if we attempt to read past EOF (to ensure
725 * i_size is up to date).
726 */
727 if (fc->auto_inval_data ||
728 (pos + iov_length(iov, nr_segs) > i_size_read(inode))) {
729 int err;
730 err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
731 if (err)
732 return err;
733 }
734
735 return generic_file_aio_read(iocb, iov, nr_segs, pos);
736 }
737
738 static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
739 loff_t pos, size_t count)
740 {
741 struct fuse_write_in *inarg = &req->misc.write.in;
742 struct fuse_write_out *outarg = &req->misc.write.out;
743
744 inarg->fh = ff->fh;
745 inarg->offset = pos;
746 inarg->size = count;
747 req->in.h.opcode = FUSE_WRITE;
748 req->in.h.nodeid = ff->nodeid;
749 req->in.numargs = 2;
750 if (ff->fc->minor < 9)
751 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
752 else
753 req->in.args[0].size = sizeof(struct fuse_write_in);
754 req->in.args[0].value = inarg;
755 req->in.args[1].size = count;
756 req->out.numargs = 1;
757 req->out.args[0].size = sizeof(struct fuse_write_out);
758 req->out.args[0].value = outarg;
759 }
760
761 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
762 loff_t pos, size_t count, fl_owner_t owner)
763 {
764 struct fuse_file *ff = file->private_data;
765 struct fuse_conn *fc = ff->fc;
766 struct fuse_write_in *inarg = &req->misc.write.in;
767
768 fuse_write_fill(req, ff, pos, count);
769 inarg->flags = file->f_flags;
770 if (owner != NULL) {
771 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
772 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
773 }
774 fuse_request_send(fc, req);
775 return req->misc.write.out.size;
776 }
777
778 void fuse_write_update_size(struct inode *inode, loff_t pos)
779 {
780 struct fuse_conn *fc = get_fuse_conn(inode);
781 struct fuse_inode *fi = get_fuse_inode(inode);
782
783 spin_lock(&fc->lock);
784 fi->attr_version = ++fc->attr_version;
785 if (pos > inode->i_size)
786 i_size_write(inode, pos);
787 spin_unlock(&fc->lock);
788 }
789
790 static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file,
791 struct inode *inode, loff_t pos,
792 size_t count)
793 {
794 size_t res;
795 unsigned offset;
796 unsigned i;
797
798 for (i = 0; i < req->num_pages; i++)
799 fuse_wait_on_page_writeback(inode, req->pages[i]->index);
800
801 res = fuse_send_write(req, file, pos, count, NULL);
802
803 offset = req->page_descs[0].offset;
804 count = res;
805 for (i = 0; i < req->num_pages; i++) {
806 struct page *page = req->pages[i];
807
808 if (!req->out.h.error && !offset && count >= PAGE_CACHE_SIZE)
809 SetPageUptodate(page);
810
811 if (count > PAGE_CACHE_SIZE - offset)
812 count -= PAGE_CACHE_SIZE - offset;
813 else
814 count = 0;
815 offset = 0;
816
817 unlock_page(page);
818 page_cache_release(page);
819 }
820
821 return res;
822 }
823
824 static ssize_t fuse_fill_write_pages(struct fuse_req *req,
825 struct address_space *mapping,
826 struct iov_iter *ii, loff_t pos)
827 {
828 struct fuse_conn *fc = get_fuse_conn(mapping->host);
829 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
830 size_t count = 0;
831 int err;
832
833 req->in.argpages = 1;
834 req->page_descs[0].offset = offset;
835
836 do {
837 size_t tmp;
838 struct page *page;
839 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
840 size_t bytes = min_t(size_t, PAGE_CACHE_SIZE - offset,
841 iov_iter_count(ii));
842
843 bytes = min_t(size_t, bytes, fc->max_write - count);
844
845 again:
846 err = -EFAULT;
847 if (iov_iter_fault_in_readable(ii, bytes))
848 break;
849
850 err = -ENOMEM;
851 page = grab_cache_page_write_begin(mapping, index, 0);
852 if (!page)
853 break;
854
855 if (mapping_writably_mapped(mapping))
856 flush_dcache_page(page);
857
858 pagefault_disable();
859 tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
860 pagefault_enable();
861 flush_dcache_page(page);
862
863 mark_page_accessed(page);
864
865 if (!tmp) {
866 unlock_page(page);
867 page_cache_release(page);
868 bytes = min(bytes, iov_iter_single_seg_count(ii));
869 goto again;
870 }
871
872 err = 0;
873 req->pages[req->num_pages] = page;
874 req->page_descs[req->num_pages].length = tmp;
875 req->num_pages++;
876
877 iov_iter_advance(ii, tmp);
878 count += tmp;
879 pos += tmp;
880 offset += tmp;
881 if (offset == PAGE_CACHE_SIZE)
882 offset = 0;
883
884 if (!fc->big_writes)
885 break;
886 } while (iov_iter_count(ii) && count < fc->max_write &&
887 req->num_pages < req->max_pages && offset == 0);
888
889 return count > 0 ? count : err;
890 }
891
892 static inline unsigned fuse_wr_pages(loff_t pos, size_t len)
893 {
894 return min_t(unsigned,
895 ((pos + len - 1) >> PAGE_CACHE_SHIFT) -
896 (pos >> PAGE_CACHE_SHIFT) + 1,
897 FUSE_MAX_PAGES_PER_REQ);
898 }
899
900 static ssize_t fuse_perform_write(struct file *file,
901 struct address_space *mapping,
902 struct iov_iter *ii, loff_t pos)
903 {
904 struct inode *inode = mapping->host;
905 struct fuse_conn *fc = get_fuse_conn(inode);
906 int err = 0;
907 ssize_t res = 0;
908
909 if (is_bad_inode(inode))
910 return -EIO;
911
912 do {
913 struct fuse_req *req;
914 ssize_t count;
915 unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii));
916
917 req = fuse_get_req(fc, nr_pages);
918 if (IS_ERR(req)) {
919 err = PTR_ERR(req);
920 break;
921 }
922
923 count = fuse_fill_write_pages(req, mapping, ii, pos);
924 if (count <= 0) {
925 err = count;
926 } else {
927 size_t num_written;
928
929 num_written = fuse_send_write_pages(req, file, inode,
930 pos, count);
931 err = req->out.h.error;
932 if (!err) {
933 res += num_written;
934 pos += num_written;
935
936 /* break out of the loop on short write */
937 if (num_written != count)
938 err = -EIO;
939 }
940 }
941 fuse_put_request(fc, req);
942 } while (!err && iov_iter_count(ii));
943
944 if (res > 0)
945 fuse_write_update_size(inode, pos);
946
947 fuse_invalidate_attr(inode);
948
949 return res > 0 ? res : err;
950 }
951
952 static ssize_t fuse_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
953 unsigned long nr_segs, loff_t pos)
954 {
955 struct file *file = iocb->ki_filp;
956 struct address_space *mapping = file->f_mapping;
957 size_t count = 0;
958 size_t ocount = 0;
959 ssize_t written = 0;
960 ssize_t written_buffered = 0;
961 struct inode *inode = mapping->host;
962 ssize_t err;
963 struct iov_iter i;
964 loff_t endbyte = 0;
965
966 WARN_ON(iocb->ki_pos != pos);
967
968 ocount = 0;
969 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
970 if (err)
971 return err;
972
973 count = ocount;
974 sb_start_write(inode->i_sb);
975 mutex_lock(&inode->i_mutex);
976
977 /* We can write back this queue in page reclaim */
978 current->backing_dev_info = mapping->backing_dev_info;
979
980 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
981 if (err)
982 goto out;
983
984 if (count == 0)
985 goto out;
986
987 err = file_remove_suid(file);
988 if (err)
989 goto out;
990
991 err = file_update_time(file);
992 if (err)
993 goto out;
994
995 if (file->f_flags & O_DIRECT) {
996 written = generic_file_direct_write(iocb, iov, &nr_segs,
997 pos, &iocb->ki_pos,
998 count, ocount);
999 if (written < 0 || written == count)
1000 goto out;
1001
1002 pos += written;
1003 count -= written;
1004
1005 iov_iter_init(&i, iov, nr_segs, count, written);
1006 written_buffered = fuse_perform_write(file, mapping, &i, pos);
1007 if (written_buffered < 0) {
1008 err = written_buffered;
1009 goto out;
1010 }
1011 endbyte = pos + written_buffered - 1;
1012
1013 err = filemap_write_and_wait_range(file->f_mapping, pos,
1014 endbyte);
1015 if (err)
1016 goto out;
1017
1018 invalidate_mapping_pages(file->f_mapping,
1019 pos >> PAGE_CACHE_SHIFT,
1020 endbyte >> PAGE_CACHE_SHIFT);
1021
1022 written += written_buffered;
1023 iocb->ki_pos = pos + written_buffered;
1024 } else {
1025 iov_iter_init(&i, iov, nr_segs, count, 0);
1026 written = fuse_perform_write(file, mapping, &i, pos);
1027 if (written >= 0)
1028 iocb->ki_pos = pos + written;
1029 }
1030 out:
1031 current->backing_dev_info = NULL;
1032 mutex_unlock(&inode->i_mutex);
1033 sb_end_write(inode->i_sb);
1034
1035 return written ? written : err;
1036 }
1037
1038 static void fuse_release_user_pages(struct fuse_req *req, int write)
1039 {
1040 unsigned i;
1041
1042 for (i = 0; i < req->num_pages; i++) {
1043 struct page *page = req->pages[i];
1044 if (write)
1045 set_page_dirty_lock(page);
1046 put_page(page);
1047 }
1048 }
1049
1050 static inline void fuse_page_descs_length_init(struct fuse_req *req)
1051 {
1052 int i;
1053
1054 for (i = 0; i < req->num_pages; i++)
1055 req->page_descs[i].length = PAGE_SIZE -
1056 req->page_descs[i].offset;
1057 }
1058
1059 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
1060 size_t *nbytesp, int write)
1061 {
1062 size_t nbytes = *nbytesp;
1063 unsigned long user_addr = (unsigned long) buf;
1064 unsigned offset = user_addr & ~PAGE_MASK;
1065 int npages;
1066
1067 /* Special case for kernel I/O: can copy directly into the buffer */
1068 if (segment_eq(get_fs(), KERNEL_DS)) {
1069 if (write)
1070 req->in.args[1].value = (void *) user_addr;
1071 else
1072 req->out.args[0].value = (void *) user_addr;
1073
1074 return 0;
1075 }
1076
1077 nbytes = min_t(size_t, nbytes, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
1078 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1079 npages = clamp(npages, 1, FUSE_MAX_PAGES_PER_REQ);
1080 npages = get_user_pages_fast(user_addr, npages, !write, req->pages);
1081 if (npages < 0)
1082 return npages;
1083
1084 req->num_pages = npages;
1085 req->page_descs[0].offset = offset;
1086 fuse_page_descs_length_init(req);
1087
1088 if (write)
1089 req->in.argpages = 1;
1090 else
1091 req->out.argpages = 1;
1092
1093 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_descs[0].offset;
1094
1095 if (*nbytesp < nbytes)
1096 req->page_descs[req->num_pages - 1].length -=
1097 nbytes - *nbytesp;
1098
1099 *nbytesp = min(*nbytesp, nbytes);
1100
1101 return 0;
1102 }
1103
1104 ssize_t fuse_direct_io(struct file *file, const char __user *buf,
1105 size_t count, loff_t *ppos, int write)
1106 {
1107 struct fuse_file *ff = file->private_data;
1108 struct fuse_conn *fc = ff->fc;
1109 size_t nmax = write ? fc->max_write : fc->max_read;
1110 loff_t pos = *ppos;
1111 ssize_t res = 0;
1112 struct fuse_req *req;
1113
1114 req = fuse_get_req(fc, FUSE_MAX_PAGES_PER_REQ);
1115 if (IS_ERR(req))
1116 return PTR_ERR(req);
1117
1118 while (count) {
1119 size_t nres;
1120 fl_owner_t owner = current->files;
1121 size_t nbytes = min(count, nmax);
1122 int err = fuse_get_user_pages(req, buf, &nbytes, write);
1123 if (err) {
1124 res = err;
1125 break;
1126 }
1127
1128 if (write)
1129 nres = fuse_send_write(req, file, pos, nbytes, owner);
1130 else
1131 nres = fuse_send_read(req, file, pos, nbytes, owner);
1132
1133 fuse_release_user_pages(req, !write);
1134 if (req->out.h.error) {
1135 if (!res)
1136 res = req->out.h.error;
1137 break;
1138 } else if (nres > nbytes) {
1139 res = -EIO;
1140 break;
1141 }
1142 count -= nres;
1143 res += nres;
1144 pos += nres;
1145 buf += nres;
1146 if (nres != nbytes)
1147 break;
1148 if (count) {
1149 fuse_put_request(fc, req);
1150 req = fuse_get_req(fc, FUSE_MAX_PAGES_PER_REQ);
1151 if (IS_ERR(req))
1152 break;
1153 }
1154 }
1155 if (!IS_ERR(req))
1156 fuse_put_request(fc, req);
1157 if (res > 0)
1158 *ppos = pos;
1159
1160 return res;
1161 }
1162 EXPORT_SYMBOL_GPL(fuse_direct_io);
1163
1164 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
1165 size_t count, loff_t *ppos)
1166 {
1167 ssize_t res;
1168 struct inode *inode = file->f_path.dentry->d_inode;
1169
1170 if (is_bad_inode(inode))
1171 return -EIO;
1172
1173 res = fuse_direct_io(file, buf, count, ppos, 0);
1174
1175 fuse_invalidate_attr(inode);
1176
1177 return res;
1178 }
1179
1180 static ssize_t __fuse_direct_write(struct file *file, const char __user *buf,
1181 size_t count, loff_t *ppos)
1182 {
1183 struct inode *inode = file->f_path.dentry->d_inode;
1184 ssize_t res;
1185
1186 res = generic_write_checks(file, ppos, &count, 0);
1187 if (!res) {
1188 res = fuse_direct_io(file, buf, count, ppos, 1);
1189 if (res > 0)
1190 fuse_write_update_size(inode, *ppos);
1191 }
1192
1193 fuse_invalidate_attr(inode);
1194
1195 return res;
1196 }
1197
1198 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
1199 size_t count, loff_t *ppos)
1200 {
1201 struct inode *inode = file->f_path.dentry->d_inode;
1202 ssize_t res;
1203
1204 if (is_bad_inode(inode))
1205 return -EIO;
1206
1207 /* Don't allow parallel writes to the same file */
1208 mutex_lock(&inode->i_mutex);
1209 res = __fuse_direct_write(file, buf, count, ppos);
1210 mutex_unlock(&inode->i_mutex);
1211
1212 return res;
1213 }
1214
1215 static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
1216 {
1217 __free_page(req->pages[0]);
1218 fuse_file_put(req->ff, false);
1219 }
1220
1221 static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
1222 {
1223 struct inode *inode = req->inode;
1224 struct fuse_inode *fi = get_fuse_inode(inode);
1225 struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info;
1226
1227 list_del(&req->writepages_entry);
1228 dec_bdi_stat(bdi, BDI_WRITEBACK);
1229 dec_zone_page_state(req->pages[0], NR_WRITEBACK_TEMP);
1230 bdi_writeout_inc(bdi);
1231 wake_up(&fi->page_waitq);
1232 }
1233
1234 /* Called under fc->lock, may release and reacquire it */
1235 static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req)
1236 __releases(fc->lock)
1237 __acquires(fc->lock)
1238 {
1239 struct fuse_inode *fi = get_fuse_inode(req->inode);
1240 loff_t size = i_size_read(req->inode);
1241 struct fuse_write_in *inarg = &req->misc.write.in;
1242
1243 if (!fc->connected)
1244 goto out_free;
1245
1246 if (inarg->offset + PAGE_CACHE_SIZE <= size) {
1247 inarg->size = PAGE_CACHE_SIZE;
1248 } else if (inarg->offset < size) {
1249 inarg->size = size & (PAGE_CACHE_SIZE - 1);
1250 } else {
1251 /* Got truncated off completely */
1252 goto out_free;
1253 }
1254
1255 req->in.args[1].size = inarg->size;
1256 fi->writectr++;
1257 fuse_request_send_background_locked(fc, req);
1258 return;
1259
1260 out_free:
1261 fuse_writepage_finish(fc, req);
1262 spin_unlock(&fc->lock);
1263 fuse_writepage_free(fc, req);
1264 fuse_put_request(fc, req);
1265 spin_lock(&fc->lock);
1266 }
1267
1268 /*
1269 * If fi->writectr is positive (no truncate or fsync going on) send
1270 * all queued writepage requests.
1271 *
1272 * Called with fc->lock
1273 */
1274 void fuse_flush_writepages(struct inode *inode)
1275 __releases(fc->lock)
1276 __acquires(fc->lock)
1277 {
1278 struct fuse_conn *fc = get_fuse_conn(inode);
1279 struct fuse_inode *fi = get_fuse_inode(inode);
1280 struct fuse_req *req;
1281
1282 while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1283 req = list_entry(fi->queued_writes.next, struct fuse_req, list);
1284 list_del_init(&req->list);
1285 fuse_send_writepage(fc, req);
1286 }
1287 }
1288
1289 static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
1290 {
1291 struct inode *inode = req->inode;
1292 struct fuse_inode *fi = get_fuse_inode(inode);
1293
1294 mapping_set_error(inode->i_mapping, req->out.h.error);
1295 spin_lock(&fc->lock);
1296 fi->writectr--;
1297 fuse_writepage_finish(fc, req);
1298 spin_unlock(&fc->lock);
1299 fuse_writepage_free(fc, req);
1300 }
1301
1302 static int fuse_writepage_locked(struct page *page)
1303 {
1304 struct address_space *mapping = page->mapping;
1305 struct inode *inode = mapping->host;
1306 struct fuse_conn *fc = get_fuse_conn(inode);
1307 struct fuse_inode *fi = get_fuse_inode(inode);
1308 struct fuse_req *req;
1309 struct fuse_file *ff;
1310 struct page *tmp_page;
1311
1312 set_page_writeback(page);
1313
1314 req = fuse_request_alloc_nofs(1);
1315 if (!req)
1316 goto err;
1317
1318 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1319 if (!tmp_page)
1320 goto err_free;
1321
1322 spin_lock(&fc->lock);
1323 BUG_ON(list_empty(&fi->write_files));
1324 ff = list_entry(fi->write_files.next, struct fuse_file, write_entry);
1325 req->ff = fuse_file_get(ff);
1326 spin_unlock(&fc->lock);
1327
1328 fuse_write_fill(req, ff, page_offset(page), 0);
1329
1330 copy_highpage(tmp_page, page);
1331 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
1332 req->in.argpages = 1;
1333 req->num_pages = 1;
1334 req->pages[0] = tmp_page;
1335 req->page_descs[0].offset = 0;
1336 req->page_descs[0].length = PAGE_SIZE;
1337 req->end = fuse_writepage_end;
1338 req->inode = inode;
1339
1340 inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK);
1341 inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP);
1342 end_page_writeback(page);
1343
1344 spin_lock(&fc->lock);
1345 list_add(&req->writepages_entry, &fi->writepages);
1346 list_add_tail(&req->list, &fi->queued_writes);
1347 fuse_flush_writepages(inode);
1348 spin_unlock(&fc->lock);
1349
1350 return 0;
1351
1352 err_free:
1353 fuse_request_free(req);
1354 err:
1355 end_page_writeback(page);
1356 return -ENOMEM;
1357 }
1358
1359 static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1360 {
1361 int err;
1362
1363 err = fuse_writepage_locked(page);
1364 unlock_page(page);
1365
1366 return err;
1367 }
1368
1369 static int fuse_launder_page(struct page *page)
1370 {
1371 int err = 0;
1372 if (clear_page_dirty_for_io(page)) {
1373 struct inode *inode = page->mapping->host;
1374 err = fuse_writepage_locked(page);
1375 if (!err)
1376 fuse_wait_on_page_writeback(inode, page->index);
1377 }
1378 return err;
1379 }
1380
1381 /*
1382 * Write back dirty pages now, because there may not be any suitable
1383 * open files later
1384 */
1385 static void fuse_vma_close(struct vm_area_struct *vma)
1386 {
1387 filemap_write_and_wait(vma->vm_file->f_mapping);
1388 }
1389
1390 /*
1391 * Wait for writeback against this page to complete before allowing it
1392 * to be marked dirty again, and hence written back again, possibly
1393 * before the previous writepage completed.
1394 *
1395 * Block here, instead of in ->writepage(), so that the userspace fs
1396 * can only block processes actually operating on the filesystem.
1397 *
1398 * Otherwise unprivileged userspace fs would be able to block
1399 * unrelated:
1400 *
1401 * - page migration
1402 * - sync(2)
1403 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
1404 */
1405 static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1406 {
1407 struct page *page = vmf->page;
1408 /*
1409 * Don't use page->mapping as it may become NULL from a
1410 * concurrent truncate.
1411 */
1412 struct inode *inode = vma->vm_file->f_mapping->host;
1413
1414 fuse_wait_on_page_writeback(inode, page->index);
1415 return 0;
1416 }
1417
1418 static const struct vm_operations_struct fuse_file_vm_ops = {
1419 .close = fuse_vma_close,
1420 .fault = filemap_fault,
1421 .page_mkwrite = fuse_page_mkwrite,
1422 .remap_pages = generic_file_remap_pages,
1423 };
1424
1425 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
1426 {
1427 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) {
1428 struct inode *inode = file->f_dentry->d_inode;
1429 struct fuse_conn *fc = get_fuse_conn(inode);
1430 struct fuse_inode *fi = get_fuse_inode(inode);
1431 struct fuse_file *ff = file->private_data;
1432 /*
1433 * file may be written through mmap, so chain it onto the
1434 * inodes's write_file list
1435 */
1436 spin_lock(&fc->lock);
1437 if (list_empty(&ff->write_entry))
1438 list_add(&ff->write_entry, &fi->write_files);
1439 spin_unlock(&fc->lock);
1440 }
1441 file_accessed(file);
1442 vma->vm_ops = &fuse_file_vm_ops;
1443 return 0;
1444 }
1445
1446 static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma)
1447 {
1448 /* Can't provide the coherency needed for MAP_SHARED */
1449 if (vma->vm_flags & VM_MAYSHARE)
1450 return -ENODEV;
1451
1452 invalidate_inode_pages2(file->f_mapping);
1453
1454 return generic_file_mmap(file, vma);
1455 }
1456
1457 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
1458 struct file_lock *fl)
1459 {
1460 switch (ffl->type) {
1461 case F_UNLCK:
1462 break;
1463
1464 case F_RDLCK:
1465 case F_WRLCK:
1466 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
1467 ffl->end < ffl->start)
1468 return -EIO;
1469
1470 fl->fl_start = ffl->start;
1471 fl->fl_end = ffl->end;
1472 fl->fl_pid = ffl->pid;
1473 break;
1474
1475 default:
1476 return -EIO;
1477 }
1478 fl->fl_type = ffl->type;
1479 return 0;
1480 }
1481
1482 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
1483 const struct file_lock *fl, int opcode, pid_t pid,
1484 int flock)
1485 {
1486 struct inode *inode = file->f_path.dentry->d_inode;
1487 struct fuse_conn *fc = get_fuse_conn(inode);
1488 struct fuse_file *ff = file->private_data;
1489 struct fuse_lk_in *arg = &req->misc.lk_in;
1490
1491 arg->fh = ff->fh;
1492 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
1493 arg->lk.start = fl->fl_start;
1494 arg->lk.end = fl->fl_end;
1495 arg->lk.type = fl->fl_type;
1496 arg->lk.pid = pid;
1497 if (flock)
1498 arg->lk_flags |= FUSE_LK_FLOCK;
1499 req->in.h.opcode = opcode;
1500 req->in.h.nodeid = get_node_id(inode);
1501 req->in.numargs = 1;
1502 req->in.args[0].size = sizeof(*arg);
1503 req->in.args[0].value = arg;
1504 }
1505
1506 static int fuse_getlk(struct file *file, struct file_lock *fl)
1507 {
1508 struct inode *inode = file->f_path.dentry->d_inode;
1509 struct fuse_conn *fc = get_fuse_conn(inode);
1510 struct fuse_req *req;
1511 struct fuse_lk_out outarg;
1512 int err;
1513
1514 req = fuse_get_req_nopages(fc);
1515 if (IS_ERR(req))
1516 return PTR_ERR(req);
1517
1518 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0);
1519 req->out.numargs = 1;
1520 req->out.args[0].size = sizeof(outarg);
1521 req->out.args[0].value = &outarg;
1522 fuse_request_send(fc, req);
1523 err = req->out.h.error;
1524 fuse_put_request(fc, req);
1525 if (!err)
1526 err = convert_fuse_file_lock(&outarg.lk, fl);
1527
1528 return err;
1529 }
1530
1531 static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
1532 {
1533 struct inode *inode = file->f_path.dentry->d_inode;
1534 struct fuse_conn *fc = get_fuse_conn(inode);
1535 struct fuse_req *req;
1536 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
1537 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
1538 int err;
1539
1540 if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
1541 /* NLM needs asynchronous locks, which we don't support yet */
1542 return -ENOLCK;
1543 }
1544
1545 /* Unlock on close is handled by the flush method */
1546 if (fl->fl_flags & FL_CLOSE)
1547 return 0;
1548
1549 req = fuse_get_req_nopages(fc);
1550 if (IS_ERR(req))
1551 return PTR_ERR(req);
1552
1553 fuse_lk_fill(req, file, fl, opcode, pid, flock);
1554 fuse_request_send(fc, req);
1555 err = req->out.h.error;
1556 /* locking is restartable */
1557 if (err == -EINTR)
1558 err = -ERESTARTSYS;
1559 fuse_put_request(fc, req);
1560 return err;
1561 }
1562
1563 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
1564 {
1565 struct inode *inode = file->f_path.dentry->d_inode;
1566 struct fuse_conn *fc = get_fuse_conn(inode);
1567 int err;
1568
1569 if (cmd == F_CANCELLK) {
1570 err = 0;
1571 } else if (cmd == F_GETLK) {
1572 if (fc->no_lock) {
1573 posix_test_lock(file, fl);
1574 err = 0;
1575 } else
1576 err = fuse_getlk(file, fl);
1577 } else {
1578 if (fc->no_lock)
1579 err = posix_lock_file(file, fl, NULL);
1580 else
1581 err = fuse_setlk(file, fl, 0);
1582 }
1583 return err;
1584 }
1585
1586 static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
1587 {
1588 struct inode *inode = file->f_path.dentry->d_inode;
1589 struct fuse_conn *fc = get_fuse_conn(inode);
1590 int err;
1591
1592 if (fc->no_flock) {
1593 err = flock_lock_file_wait(file, fl);
1594 } else {
1595 struct fuse_file *ff = file->private_data;
1596
1597 /* emulate flock with POSIX locks */
1598 fl->fl_owner = (fl_owner_t) file;
1599 ff->flock = true;
1600 err = fuse_setlk(file, fl, 1);
1601 }
1602
1603 return err;
1604 }
1605
1606 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
1607 {
1608 struct inode *inode = mapping->host;
1609 struct fuse_conn *fc = get_fuse_conn(inode);
1610 struct fuse_req *req;
1611 struct fuse_bmap_in inarg;
1612 struct fuse_bmap_out outarg;
1613 int err;
1614
1615 if (!inode->i_sb->s_bdev || fc->no_bmap)
1616 return 0;
1617
1618 req = fuse_get_req_nopages(fc);
1619 if (IS_ERR(req))
1620 return 0;
1621
1622 memset(&inarg, 0, sizeof(inarg));
1623 inarg.block = block;
1624 inarg.blocksize = inode->i_sb->s_blocksize;
1625 req->in.h.opcode = FUSE_BMAP;
1626 req->in.h.nodeid = get_node_id(inode);
1627 req->in.numargs = 1;
1628 req->in.args[0].size = sizeof(inarg);
1629 req->in.args[0].value = &inarg;
1630 req->out.numargs = 1;
1631 req->out.args[0].size = sizeof(outarg);
1632 req->out.args[0].value = &outarg;
1633 fuse_request_send(fc, req);
1634 err = req->out.h.error;
1635 fuse_put_request(fc, req);
1636 if (err == -ENOSYS)
1637 fc->no_bmap = 1;
1638
1639 return err ? 0 : outarg.block;
1640 }
1641
1642 static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence)
1643 {
1644 loff_t retval;
1645 struct inode *inode = file->f_path.dentry->d_inode;
1646
1647 /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
1648 if (whence == SEEK_CUR || whence == SEEK_SET)
1649 return generic_file_llseek(file, offset, whence);
1650
1651 mutex_lock(&inode->i_mutex);
1652 retval = fuse_update_attributes(inode, NULL, file, NULL);
1653 if (!retval)
1654 retval = generic_file_llseek(file, offset, whence);
1655 mutex_unlock(&inode->i_mutex);
1656
1657 return retval;
1658 }
1659
1660 static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov,
1661 unsigned int nr_segs, size_t bytes, bool to_user)
1662 {
1663 struct iov_iter ii;
1664 int page_idx = 0;
1665
1666 if (!bytes)
1667 return 0;
1668
1669 iov_iter_init(&ii, iov, nr_segs, bytes, 0);
1670
1671 while (iov_iter_count(&ii)) {
1672 struct page *page = pages[page_idx++];
1673 size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii));
1674 void *kaddr;
1675
1676 kaddr = kmap(page);
1677
1678 while (todo) {
1679 char __user *uaddr = ii.iov->iov_base + ii.iov_offset;
1680 size_t iov_len = ii.iov->iov_len - ii.iov_offset;
1681 size_t copy = min(todo, iov_len);
1682 size_t left;
1683
1684 if (!to_user)
1685 left = copy_from_user(kaddr, uaddr, copy);
1686 else
1687 left = copy_to_user(uaddr, kaddr, copy);
1688
1689 if (unlikely(left))
1690 return -EFAULT;
1691
1692 iov_iter_advance(&ii, copy);
1693 todo -= copy;
1694 kaddr += copy;
1695 }
1696
1697 kunmap(page);
1698 }
1699
1700 return 0;
1701 }
1702
1703 /*
1704 * CUSE servers compiled on 32bit broke on 64bit kernels because the
1705 * ABI was defined to be 'struct iovec' which is different on 32bit
1706 * and 64bit. Fortunately we can determine which structure the server
1707 * used from the size of the reply.
1708 */
1709 static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
1710 size_t transferred, unsigned count,
1711 bool is_compat)
1712 {
1713 #ifdef CONFIG_COMPAT
1714 if (count * sizeof(struct compat_iovec) == transferred) {
1715 struct compat_iovec *ciov = src;
1716 unsigned i;
1717
1718 /*
1719 * With this interface a 32bit server cannot support
1720 * non-compat (i.e. ones coming from 64bit apps) ioctl
1721 * requests
1722 */
1723 if (!is_compat)
1724 return -EINVAL;
1725
1726 for (i = 0; i < count; i++) {
1727 dst[i].iov_base = compat_ptr(ciov[i].iov_base);
1728 dst[i].iov_len = ciov[i].iov_len;
1729 }
1730 return 0;
1731 }
1732 #endif
1733
1734 if (count * sizeof(struct iovec) != transferred)
1735 return -EIO;
1736
1737 memcpy(dst, src, transferred);
1738 return 0;
1739 }
1740
1741 /* Make sure iov_length() won't overflow */
1742 static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count)
1743 {
1744 size_t n;
1745 u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT;
1746
1747 for (n = 0; n < count; n++, iov++) {
1748 if (iov->iov_len > (size_t) max)
1749 return -ENOMEM;
1750 max -= iov->iov_len;
1751 }
1752 return 0;
1753 }
1754
1755 static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
1756 void *src, size_t transferred, unsigned count,
1757 bool is_compat)
1758 {
1759 unsigned i;
1760 struct fuse_ioctl_iovec *fiov = src;
1761
1762 if (fc->minor < 16) {
1763 return fuse_copy_ioctl_iovec_old(dst, src, transferred,
1764 count, is_compat);
1765 }
1766
1767 if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
1768 return -EIO;
1769
1770 for (i = 0; i < count; i++) {
1771 /* Did the server supply an inappropriate value? */
1772 if (fiov[i].base != (unsigned long) fiov[i].base ||
1773 fiov[i].len != (unsigned long) fiov[i].len)
1774 return -EIO;
1775
1776 dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
1777 dst[i].iov_len = (size_t) fiov[i].len;
1778
1779 #ifdef CONFIG_COMPAT
1780 if (is_compat &&
1781 (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
1782 (compat_size_t) dst[i].iov_len != fiov[i].len))
1783 return -EIO;
1784 #endif
1785 }
1786
1787 return 0;
1788 }
1789
1790
1791 /*
1792 * For ioctls, there is no generic way to determine how much memory
1793 * needs to be read and/or written. Furthermore, ioctls are allowed
1794 * to dereference the passed pointer, so the parameter requires deep
1795 * copying but FUSE has no idea whatsoever about what to copy in or
1796 * out.
1797 *
1798 * This is solved by allowing FUSE server to retry ioctl with
1799 * necessary in/out iovecs. Let's assume the ioctl implementation
1800 * needs to read in the following structure.
1801 *
1802 * struct a {
1803 * char *buf;
1804 * size_t buflen;
1805 * }
1806 *
1807 * On the first callout to FUSE server, inarg->in_size and
1808 * inarg->out_size will be NULL; then, the server completes the ioctl
1809 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
1810 * the actual iov array to
1811 *
1812 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
1813 *
1814 * which tells FUSE to copy in the requested area and retry the ioctl.
1815 * On the second round, the server has access to the structure and
1816 * from that it can tell what to look for next, so on the invocation,
1817 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
1818 *
1819 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
1820 * { .iov_base = a.buf, .iov_len = a.buflen } }
1821 *
1822 * FUSE will copy both struct a and the pointed buffer from the
1823 * process doing the ioctl and retry ioctl with both struct a and the
1824 * buffer.
1825 *
1826 * This time, FUSE server has everything it needs and completes ioctl
1827 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
1828 *
1829 * Copying data out works the same way.
1830 *
1831 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
1832 * automatically initializes in and out iovs by decoding @cmd with
1833 * _IOC_* macros and the server is not allowed to request RETRY. This
1834 * limits ioctl data transfers to well-formed ioctls and is the forced
1835 * behavior for all FUSE servers.
1836 */
1837 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1838 unsigned int flags)
1839 {
1840 struct fuse_file *ff = file->private_data;
1841 struct fuse_conn *fc = ff->fc;
1842 struct fuse_ioctl_in inarg = {
1843 .fh = ff->fh,
1844 .cmd = cmd,
1845 .arg = arg,
1846 .flags = flags
1847 };
1848 struct fuse_ioctl_out outarg;
1849 struct fuse_req *req = NULL;
1850 struct page **pages = NULL;
1851 struct iovec *iov_page = NULL;
1852 struct iovec *in_iov = NULL, *out_iov = NULL;
1853 unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages;
1854 size_t in_size, out_size, transferred;
1855 int err;
1856
1857 #if BITS_PER_LONG == 32
1858 inarg.flags |= FUSE_IOCTL_32BIT;
1859 #else
1860 if (flags & FUSE_IOCTL_COMPAT)
1861 inarg.flags |= FUSE_IOCTL_32BIT;
1862 #endif
1863
1864 /* assume all the iovs returned by client always fits in a page */
1865 BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
1866
1867 err = -ENOMEM;
1868 pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL);
1869 iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
1870 if (!pages || !iov_page)
1871 goto out;
1872
1873 /*
1874 * If restricted, initialize IO parameters as encoded in @cmd.
1875 * RETRY from server is not allowed.
1876 */
1877 if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
1878 struct iovec *iov = iov_page;
1879
1880 iov->iov_base = (void __user *)arg;
1881 iov->iov_len = _IOC_SIZE(cmd);
1882
1883 if (_IOC_DIR(cmd) & _IOC_WRITE) {
1884 in_iov = iov;
1885 in_iovs = 1;
1886 }
1887
1888 if (_IOC_DIR(cmd) & _IOC_READ) {
1889 out_iov = iov;
1890 out_iovs = 1;
1891 }
1892 }
1893
1894 retry:
1895 inarg.in_size = in_size = iov_length(in_iov, in_iovs);
1896 inarg.out_size = out_size = iov_length(out_iov, out_iovs);
1897
1898 /*
1899 * Out data can be used either for actual out data or iovs,
1900 * make sure there always is at least one page.
1901 */
1902 out_size = max_t(size_t, out_size, PAGE_SIZE);
1903 max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
1904
1905 /* make sure there are enough buffer pages and init request with them */
1906 err = -ENOMEM;
1907 if (max_pages > FUSE_MAX_PAGES_PER_REQ)
1908 goto out;
1909 while (num_pages < max_pages) {
1910 pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
1911 if (!pages[num_pages])
1912 goto out;
1913 num_pages++;
1914 }
1915
1916 req = fuse_get_req(fc, num_pages);
1917 if (IS_ERR(req)) {
1918 err = PTR_ERR(req);
1919 req = NULL;
1920 goto out;
1921 }
1922 memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
1923 req->num_pages = num_pages;
1924 fuse_page_descs_length_init(req);
1925
1926 /* okay, let's send it to the client */
1927 req->in.h.opcode = FUSE_IOCTL;
1928 req->in.h.nodeid = ff->nodeid;
1929 req->in.numargs = 1;
1930 req->in.args[0].size = sizeof(inarg);
1931 req->in.args[0].value = &inarg;
1932 if (in_size) {
1933 req->in.numargs++;
1934 req->in.args[1].size = in_size;
1935 req->in.argpages = 1;
1936
1937 err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size,
1938 false);
1939 if (err)
1940 goto out;
1941 }
1942
1943 req->out.numargs = 2;
1944 req->out.args[0].size = sizeof(outarg);
1945 req->out.args[0].value = &outarg;
1946 req->out.args[1].size = out_size;
1947 req->out.argpages = 1;
1948 req->out.argvar = 1;
1949
1950 fuse_request_send(fc, req);
1951 err = req->out.h.error;
1952 transferred = req->out.args[1].size;
1953 fuse_put_request(fc, req);
1954 req = NULL;
1955 if (err)
1956 goto out;
1957
1958 /* did it ask for retry? */
1959 if (outarg.flags & FUSE_IOCTL_RETRY) {
1960 void *vaddr;
1961
1962 /* no retry if in restricted mode */
1963 err = -EIO;
1964 if (!(flags & FUSE_IOCTL_UNRESTRICTED))
1965 goto out;
1966
1967 in_iovs = outarg.in_iovs;
1968 out_iovs = outarg.out_iovs;
1969
1970 /*
1971 * Make sure things are in boundary, separate checks
1972 * are to protect against overflow.
1973 */
1974 err = -ENOMEM;
1975 if (in_iovs > FUSE_IOCTL_MAX_IOV ||
1976 out_iovs > FUSE_IOCTL_MAX_IOV ||
1977 in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
1978 goto out;
1979
1980 vaddr = kmap_atomic(pages[0]);
1981 err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr,
1982 transferred, in_iovs + out_iovs,
1983 (flags & FUSE_IOCTL_COMPAT) != 0);
1984 kunmap_atomic(vaddr);
1985 if (err)
1986 goto out;
1987
1988 in_iov = iov_page;
1989 out_iov = in_iov + in_iovs;
1990
1991 err = fuse_verify_ioctl_iov(in_iov, in_iovs);
1992 if (err)
1993 goto out;
1994
1995 err = fuse_verify_ioctl_iov(out_iov, out_iovs);
1996 if (err)
1997 goto out;
1998
1999 goto retry;
2000 }
2001
2002 err = -EIO;
2003 if (transferred > inarg.out_size)
2004 goto out;
2005
2006 err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true);
2007 out:
2008 if (req)
2009 fuse_put_request(fc, req);
2010 free_page((unsigned long) iov_page);
2011 while (num_pages)
2012 __free_page(pages[--num_pages]);
2013 kfree(pages);
2014
2015 return err ? err : outarg.result;
2016 }
2017 EXPORT_SYMBOL_GPL(fuse_do_ioctl);
2018
2019 long fuse_ioctl_common(struct file *file, unsigned int cmd,
2020 unsigned long arg, unsigned int flags)
2021 {
2022 struct inode *inode = file->f_dentry->d_inode;
2023 struct fuse_conn *fc = get_fuse_conn(inode);
2024
2025 if (!fuse_allow_task(fc, current))
2026 return -EACCES;
2027
2028 if (is_bad_inode(inode))
2029 return -EIO;
2030
2031 return fuse_do_ioctl(file, cmd, arg, flags);
2032 }
2033
2034 static long fuse_file_ioctl(struct file *file, unsigned int cmd,
2035 unsigned long arg)
2036 {
2037 return fuse_ioctl_common(file, cmd, arg, 0);
2038 }
2039
2040 static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
2041 unsigned long arg)
2042 {
2043 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
2044 }
2045
2046 /*
2047 * All files which have been polled are linked to RB tree
2048 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
2049 * find the matching one.
2050 */
2051 static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
2052 struct rb_node **parent_out)
2053 {
2054 struct rb_node **link = &fc->polled_files.rb_node;
2055 struct rb_node *last = NULL;
2056
2057 while (*link) {
2058 struct fuse_file *ff;
2059
2060 last = *link;
2061 ff = rb_entry(last, struct fuse_file, polled_node);
2062
2063 if (kh < ff->kh)
2064 link = &last->rb_left;
2065 else if (kh > ff->kh)
2066 link = &last->rb_right;
2067 else
2068 return link;
2069 }
2070
2071 if (parent_out)
2072 *parent_out = last;
2073 return link;
2074 }
2075
2076 /*
2077 * The file is about to be polled. Make sure it's on the polled_files
2078 * RB tree. Note that files once added to the polled_files tree are
2079 * not removed before the file is released. This is because a file
2080 * polled once is likely to be polled again.
2081 */
2082 static void fuse_register_polled_file(struct fuse_conn *fc,
2083 struct fuse_file *ff)
2084 {
2085 spin_lock(&fc->lock);
2086 if (RB_EMPTY_NODE(&ff->polled_node)) {
2087 struct rb_node **link, *parent;
2088
2089 link = fuse_find_polled_node(fc, ff->kh, &parent);
2090 BUG_ON(*link);
2091 rb_link_node(&ff->polled_node, parent, link);
2092 rb_insert_color(&ff->polled_node, &fc->polled_files);
2093 }
2094 spin_unlock(&fc->lock);
2095 }
2096
2097 unsigned fuse_file_poll(struct file *file, poll_table *wait)
2098 {
2099 struct fuse_file *ff = file->private_data;
2100 struct fuse_conn *fc = ff->fc;
2101 struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
2102 struct fuse_poll_out outarg;
2103 struct fuse_req *req;
2104 int err;
2105
2106 if (fc->no_poll)
2107 return DEFAULT_POLLMASK;
2108
2109 poll_wait(file, &ff->poll_wait, wait);
2110
2111 /*
2112 * Ask for notification iff there's someone waiting for it.
2113 * The client may ignore the flag and always notify.
2114 */
2115 if (waitqueue_active(&ff->poll_wait)) {
2116 inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
2117 fuse_register_polled_file(fc, ff);
2118 }
2119
2120 req = fuse_get_req_nopages(fc);
2121 if (IS_ERR(req))
2122 return POLLERR;
2123
2124 req->in.h.opcode = FUSE_POLL;
2125 req->in.h.nodeid = ff->nodeid;
2126 req->in.numargs = 1;
2127 req->in.args[0].size = sizeof(inarg);
2128 req->in.args[0].value = &inarg;
2129 req->out.numargs = 1;
2130 req->out.args[0].size = sizeof(outarg);
2131 req->out.args[0].value = &outarg;
2132 fuse_request_send(fc, req);
2133 err = req->out.h.error;
2134 fuse_put_request(fc, req);
2135
2136 if (!err)
2137 return outarg.revents;
2138 if (err == -ENOSYS) {
2139 fc->no_poll = 1;
2140 return DEFAULT_POLLMASK;
2141 }
2142 return POLLERR;
2143 }
2144 EXPORT_SYMBOL_GPL(fuse_file_poll);
2145
2146 /*
2147 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
2148 * wakes up the poll waiters.
2149 */
2150 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
2151 struct fuse_notify_poll_wakeup_out *outarg)
2152 {
2153 u64 kh = outarg->kh;
2154 struct rb_node **link;
2155
2156 spin_lock(&fc->lock);
2157
2158 link = fuse_find_polled_node(fc, kh, NULL);
2159 if (*link) {
2160 struct fuse_file *ff;
2161
2162 ff = rb_entry(*link, struct fuse_file, polled_node);
2163 wake_up_interruptible_sync(&ff->poll_wait);
2164 }
2165
2166 spin_unlock(&fc->lock);
2167 return 0;
2168 }
2169
2170 static ssize_t fuse_loop_dio(struct file *filp, const struct iovec *iov,
2171 unsigned long nr_segs, loff_t *ppos, int rw)
2172 {
2173 const struct iovec *vector = iov;
2174 ssize_t ret = 0;
2175
2176 while (nr_segs > 0) {
2177 void __user *base;
2178 size_t len;
2179 ssize_t nr;
2180
2181 base = vector->iov_base;
2182 len = vector->iov_len;
2183 vector++;
2184 nr_segs--;
2185
2186 if (rw == WRITE)
2187 nr = __fuse_direct_write(filp, base, len, ppos);
2188 else
2189 nr = fuse_direct_read(filp, base, len, ppos);
2190
2191 if (nr < 0) {
2192 if (!ret)
2193 ret = nr;
2194 break;
2195 }
2196 ret += nr;
2197 if (nr != len)
2198 break;
2199 }
2200
2201 return ret;
2202 }
2203
2204
2205 static ssize_t
2206 fuse_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2207 loff_t offset, unsigned long nr_segs)
2208 {
2209 ssize_t ret = 0;
2210 struct file *file = NULL;
2211 loff_t pos = 0;
2212
2213 file = iocb->ki_filp;
2214 pos = offset;
2215
2216 ret = fuse_loop_dio(file, iov, nr_segs, &pos, rw);
2217
2218 return ret;
2219 }
2220
2221 static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
2222 loff_t length)
2223 {
2224 struct fuse_file *ff = file->private_data;
2225 struct fuse_conn *fc = ff->fc;
2226 struct fuse_req *req;
2227 struct fuse_fallocate_in inarg = {
2228 .fh = ff->fh,
2229 .offset = offset,
2230 .length = length,
2231 .mode = mode
2232 };
2233 int err;
2234
2235 if (fc->no_fallocate)
2236 return -EOPNOTSUPP;
2237
2238 req = fuse_get_req_nopages(fc);
2239 if (IS_ERR(req))
2240 return PTR_ERR(req);
2241
2242 req->in.h.opcode = FUSE_FALLOCATE;
2243 req->in.h.nodeid = ff->nodeid;
2244 req->in.numargs = 1;
2245 req->in.args[0].size = sizeof(inarg);
2246 req->in.args[0].value = &inarg;
2247 fuse_request_send(fc, req);
2248 err = req->out.h.error;
2249 if (err == -ENOSYS) {
2250 fc->no_fallocate = 1;
2251 err = -EOPNOTSUPP;
2252 }
2253 fuse_put_request(fc, req);
2254
2255 return err;
2256 }
2257
2258 static const struct file_operations fuse_file_operations = {
2259 .llseek = fuse_file_llseek,
2260 .read = do_sync_read,
2261 .aio_read = fuse_file_aio_read,
2262 .write = do_sync_write,
2263 .aio_write = fuse_file_aio_write,
2264 .mmap = fuse_file_mmap,
2265 .open = fuse_open,
2266 .flush = fuse_flush,
2267 .release = fuse_release,
2268 .fsync = fuse_fsync,
2269 .lock = fuse_file_lock,
2270 .flock = fuse_file_flock,
2271 .splice_read = generic_file_splice_read,
2272 .unlocked_ioctl = fuse_file_ioctl,
2273 .compat_ioctl = fuse_file_compat_ioctl,
2274 .poll = fuse_file_poll,
2275 .fallocate = fuse_file_fallocate,
2276 };
2277
2278 static const struct file_operations fuse_direct_io_file_operations = {
2279 .llseek = fuse_file_llseek,
2280 .read = fuse_direct_read,
2281 .write = fuse_direct_write,
2282 .mmap = fuse_direct_mmap,
2283 .open = fuse_open,
2284 .flush = fuse_flush,
2285 .release = fuse_release,
2286 .fsync = fuse_fsync,
2287 .lock = fuse_file_lock,
2288 .flock = fuse_file_flock,
2289 .unlocked_ioctl = fuse_file_ioctl,
2290 .compat_ioctl = fuse_file_compat_ioctl,
2291 .poll = fuse_file_poll,
2292 .fallocate = fuse_file_fallocate,
2293 /* no splice_read */
2294 };
2295
2296 static const struct address_space_operations fuse_file_aops = {
2297 .readpage = fuse_readpage,
2298 .writepage = fuse_writepage,
2299 .launder_page = fuse_launder_page,
2300 .readpages = fuse_readpages,
2301 .set_page_dirty = __set_page_dirty_nobuffers,
2302 .bmap = fuse_bmap,
2303 .direct_IO = fuse_direct_IO,
2304 };
2305
2306 void fuse_init_file_inode(struct inode *inode)
2307 {
2308 inode->i_fop = &fuse_file_operations;
2309 inode->i_data.a_ops = &fuse_file_aops;
2310 }
Linux kernel - Deliverables
This page took 0.235434 seconds and 5 git commands to generate.